NVIDIA Healthcare Physical AI: Cosmos, Groot H, and Isaac for Healthcare

This presentation outlines NVIDIA's strategy for Physical AI in healthcare, focusing on the integration of world foundation models, robotic policy models, and physics simulation to advance surgical robotics and hospital automation. It details the release of open-source tools like OpenAge data, Groot H, and the Cosmos suite, demonstrating how synthetic data generation accelerates the development cycle from simulation to real-world deployment.

■ Core Technology Pillars
- World Foundation Models (Cosmos) for understanding complex medical environments and generating synthetic data
- Robotic Policy Models (Groot H) enabling generalizable skills and translation from perception to action
- Physics Simulation using high-fidelity engines for training and evaluating policies before real-world testing

■ Data and Model Ecosystem
- OpenAge dataset providing over 750 hours of healthcare robotics data from multiple partners
- Groot H model architecture combining vision-language reasoning with diffusion transformers for action policies
- Integration of raw-to-insights models for ultrasound and MRI, processing data directly from sensor physics

■ Isaac for Healthcare Platform
- Domain-specific extensions including sensor simulation libraries for X-ray and ultrasound
- Synthetic data generation pipelines creating digital twins from patient CT/MR scans
- End-to-end workflows for subtask automation in surgical robotics and hospital logistics

■ Deployment and Hardware
- Holoscan 4.0 SDK supporting EtherCAT and low-latency GPU resident graphs
- IGX Thor edge AI platform providing industrial-grade compute for real-time safety-critical applications
- Hybrid approaches combining classical and neural simulation to reduce development time and cost

Developers, researchers, and healthcare technology providers can utilize these open resources to accelerate innovation in physical AI, gaining access to state-of-the-art models and simulation tools for building robust medical robotics systems.